# Prath Mode

> Route user-invoked work to the owning skill in prathamdby/skills, including multi-step delivery workflows.

- Skill: `prathamdby/prath-mode` (Agent Skill)
- Install (CLI): `npx get-skill add prathamdby/prath-mode`
- Raw SKILL.md: https://api.skillmd.com/api/skills/prathamdby/prath-mode/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: prathamdby (https://skillmd.com/u/prathamdby)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/prathamdby/prath-mode

---


# Prath mode

The leaf owns its triggers, flags, procedure, and terminal states. Read it
before acting. Never recreate a missing leaf or copy its procedure here.

## Routing map

| Immediate action                                           | Leaf                                        |
| ---------------------------------------------------------- | ------------------------------------------- |
| Commit scoped changes                                      | `commit` (`../commit/SKILL.md`)             |
| Remove code slop                                           | `deslop` (`../deslop/SKILL.md`)             |
| Create or update a PR                                      | `make-pr` (`../make-pr/SKILL.md`)           |
| Address PR feedback                                        | `fix-pr` (`../fix-pr/SKILL.md`)             |
| Inspect a PR, read threads, why CI is red, or post a reply | `gh` (`../gh/SKILL.md`)                     |
| Review a plan, design, implementation, or proposal         | `peer-review` (`../peer-review/SKILL.md`)   |
| Select among candidates or score progress                  | `verify` (`../verify/SKILL.md`)             |
| Explain a diff as HTML                                     | `explain-diff` (`../explain-diff/SKILL.md`) |
| Map or refresh the current repo                            | `recon` (`../recon/SKILL.md`)               |
| Clone or search an external repo                           | `box` (`../box/SKILL.md`)                   |
| Coordinate current-harness subagents                       | `orchestrate` (`../orchestrate/SKILL.md`)   |
| Save or resume session state                               | `handoff` (`../handoff/SKILL.md`)           |

For one action, route to its leaf. Use `orchestrate` for several in-harness
delegates. Chain only for a complete terminal outcome; several explicit
outcomes select the matching chain. A review-shaped ask stays on
`peer-review`. Do not prepend it to Ship planned work.
Orientation or reply-only terminals go to `gh`; any fix, push, or "handle
review feedback" stays `fix-pr`. `fix-pr` loads `gh` for GitHub I/O. `/gh`
never continues into `fix-pr`.

## Workflow chains

| Requested outcome                 | Ordered owners                                                   | Complete when                                 |
| --------------------------------- | ---------------------------------------------------------------- | --------------------------------------------- |
| Ship planned work                 | implementation → `deslop` → `commit` → `make-pr`                 | planned work diff tested and PR URL verified  |
| Save current work                 | optional `deslop` → `commit`                                     | new commit verified                           |
| Finish PR feedback                | `fix-pr`                                                         | `fix-pr` report complete                      |
| Inspect PR or CI, or post a reply | `gh`                                                             | script report or reply URL verified           |
| Understand current repo           | `recon`                                                          | memory and report verified                    |
| Research external code            | `box`                                                            | cited answer returned                         |
| End or resume work                | `handoff`                                                        | create or resume terminal state               |

Implementation is normal agent work, not a leaf. `fix-pr` already owns its
fix, commit, push, re-hunt, and reply loop; never append those actions.
`deslop` is required in Ship planned work and optional in custom or Save chains.
Resolve mixed staged/unstaged paths before it. Implementation is done when the
planned work's diff and relevant tests are recorded.

## 1. Match

Record a run ledger:

`route | current owner | completed owners | diff/tests | terminal`

If no route matches, ask one question about the intended outcome. Done when one
leaf or chain and its terminal condition are recorded.

## 2. Verify installation

Resolve each leaf path relative to this file and verify it exists before the
chain starts and before its turn. If any are missing, report every missing name
and checked path, then stop with:
`npx skills@latest add prathamdby/skills`

Done when all required paths exist or the missing-skill report is sent.

## 3. Invoke and resume

Read the current leaf in full and run it to one of its terminal states. Advance
only after success or no-op; pause the chain on blocked or waiting. After an
interruption, verify the last owner's artifacts before continuing.

Before `make-pr`, require a clean tree. For implementation, verify the
planned work diff and test evidence in the ledger.

Done when the recorded chain terminal condition is observed or the current
leaf has reported why progress paused.

